Yali Bian
Senior Machine Learning Engineer at Pinterest
I build and operate large-scale recommendation and ranking systems. At Pinterest I lead Light Ranking on Homefeed — the intermediate stage that scores high-volume candidates before final ranking — spanning ranking and retrieval models, serving infrastructure, and training data. Previously a Senior Research Scientist at Intel Labs; Ph.D. in Computer Science from Virginia Tech.
Experience
Sep 2025 — Present
Senior Machine Learning Engineer · Homefeed
- Tech Lead for Light Ranking, the intermediate ranking stage that scores high-volume retrieved candidates before final ranking.
- Scaled Light Ranking to ~90% of retrieved candidates by driving adoption across major sources — two-tower retrieval and Shopping/Production candidate generation.
- Improved ranking quality by training on both impressed and unimpressed examples, capturing engagement and non-engagement signals.
- Migrated Light Ranking serving from CPU to GPU, moving 100% of production traffic over and improving scalability for high-throughput scoring.
- Applied knowledge distillation to transfer signal from larger teacher models into the production ranking model.
Recommender Systems · Ranking · Two-Tower Retrieval · GPU Serving · Distillation
2022 — 2025
Senior Research Scientist, AI/ML · Human & AI Systems Research Lab
- Built recommender models addressing ranking, re-ranking, and cold-start.
- Developed deepfake-detection and explainable-AI methods using lip-reading and GAN-based techniques.
- Delivered video object segmentation and detection systems with SAM, semi-supervised object detection, CLIP, and YOLO.
PyTorch · Transformers · Vision-Language Models · Applied ML
2021 — 2022
Senior ML Infrastructure Engineer
- Designed a scalable feature lifecycle management system spanning preparation, validation, versioning, sharing, and monitoring, grounded in MLOps and distributed training.
- Built a self-serve platform for ML engineers and data scientists to manage features, models, and workflows, integrating MLflow, SageMaker, Ray, Spark, Airflow, and a Feature Store.
MLOps · Ray · MLflow · SageMaker · Spark · Airflow · Feature Store
Skills
Recommender Systems & Ranking
Candidate Generation · Embedding-Based Retrieval · Two-Tower Models · Learning to Rank · Multi-Stage Ranking · Knowledge Distillation · Cold-Start
Machine Learning & Deep Learning
PyTorch · TensorFlow · Transformers · BERT · GBDT · Representation Learning · Self-Supervised Learning
Computer Vision
Object Detection · Segmentation · Vision-Language Models · CLIP · SAM · YOLO · Deepfake Detection
ML Infrastructure & Serving
Distributed Training · GPU Inference · Low-Latency Serving · Feature Store · Model Monitoring · MLflow · SageMaker · Ray
Big Data & Pipelines
Spark · Hive · Presto · Kafka · Snowflake · Databricks · Airflow
Platforms & Languages
Kubernetes · Docker · AWS · Python · Java · C++ · Go · Scala · SQL
Education
Ph.D., Computer Science
Virginia Tech
2016 — 2022
M.S., Computer Science & Technology
Zhejiang University
2013 — 2016